feat: reporting back installation and upgrade metrics

Makefile

@@ -10,7 +10,7 @@ TERRAFORM_VERSION = 1.5.4
 OPENEBS_VERSION = 3.7.0
 K0S_VERSION = v1.27.5+k0s.0
 TROUBLESHOOT_VERSION = v0.72.0
-LD_FLAGS = -X github.com/replicatedhq/helmvm/pkg/defaults.K0sVersion=$(K0S_VERSION) -X main.Version=$(VERSION)
+LD_FLAGS = -X github.com/replicatedhq/helmvm/pkg/defaults.K0sVersion=$(K0S_VERSION) -X github.com/replicatedhq/helmvm/pkg/defaults.Version=$(VERSION)
 
 default: helmvm-linux-amd64
 

cmd/helmvm/install.go

@@ -24,6 +24,7 @@ import (
 	"github.com/replicatedhq/helmvm/pkg/defaults"
 	"github.com/replicatedhq/helmvm/pkg/goods"
 	"github.com/replicatedhq/helmvm/pkg/infra"
+	"github.com/replicatedhq/helmvm/pkg/metrics"
 	"github.com/replicatedhq/helmvm/pkg/preflights"
 	pb "github.com/replicatedhq/helmvm/pkg/progressbar"
 	"github.com/replicatedhq/helmvm/pkg/prompts"
@@ -396,33 +397,43 @@ var installCommand = &cli.Command{
 		},
 	},
 	Action: func(c *cli.Context) error {
+		metrics.ReportApplyStarted(c)
 		if defaults.DecentralizedInstall() {
 			fmt.Println("Decentralized install was detected. To manage the cluster")
 			fmt.Printf("you have to use the '%s node' commands instead.\n", defaults.BinaryName())
 			fmt.Printf("Run '%s node --help' for more information.\n", defaults.BinaryName())
+			metrics.ReportApplyFinished(c, fmt.Errorf("wrong upgrade on decentralized install"))
 			return fmt.Errorf("decentralized install detected")
 		}
 		useprompt := !c.Bool("no-prompt")
 		logrus.Infof("Materializing binaries")
 		if err := goods.Materialize(); err != nil {
-			return fmt.Errorf("unable to materialize binaries: %w", err)
+			err := fmt.Errorf("unable to materialize binaries: %w", err)
+			metrics.ReportApplyFinished(c, err)
+			return err
 		}
 		if !c.Bool("addons-only") {
 			var err error
 			var nodes []infra.Node
 			if dir := c.String("infra"); dir != "" {
 				logrus.Infof("Processing infrastructure manifests")
 				if nodes, err = infra.Apply(c.Context, dir, useprompt); err != nil {
-					return fmt.Errorf("unable to create infra: %w", err)
+					err := fmt.Errorf("unable to create infra: %w", err)
+					metrics.ReportApplyFinished(c, err)
+					return err
 				}
 			}
 			if err := applyK0sctl(c, useprompt, nodes); err != nil {
-				return fmt.Errorf("unable update cluster: %w", err)
+				err := fmt.Errorf("unable update cluster: %w", err)
+				metrics.ReportApplyFinished(c, err)
+				return err
 			}
 		}
 		logrus.Infof("Reading cluster access configuration")
 		if err := runK0sctlKubeconfig(c.Context); err != nil {
-			return fmt.Errorf("unable to get kubeconfig: %w", err)
+			err := fmt.Errorf("unable to get kubeconfig: %w", err)
+			metrics.ReportApplyFinished(c, err)
+			return err
 		}
 		logrus.Infof("Applying add-ons")
 		ccfg := defaults.PathToConfig("k0sctl.yaml")
@@ -436,16 +447,21 @@ var installCommand = &cli.Command{
 			opts = append(opts, addons.WithoutAddon(addon))
 		}
 		if err := addons.NewApplier(opts...).Apply(c.Context); err != nil {
-			return fmt.Errorf("unable to apply addons: %w", err)
+			err := fmt.Errorf("unable to apply addons: %w", err)
+			metrics.ReportApplyFinished(c, err)
+			return err
 		}
 		if err := runPostApply(c.Context); err != nil {
-			return fmt.Errorf("unable to run post apply: %w", err)
+			err := fmt.Errorf("unable to run post apply: %w", err)
+			metrics.ReportApplyFinished(c, err)
+			return err
 		}
 		fmt.Println("Cluster configuration has been applied")
 		fmt.Printf("Kubeconfig file has been placed at at %s\n", kcfg)
 		fmt.Printf("Cluster configuration file has been placed at %s\n", ccfg)
 		fmt.Println("You can now access your cluster with kubectl by running:")
 		fmt.Printf("  %s shell\n", os.Args[0])
+		metrics.ReportApplyFinished(c, nil)
 		return nil
 	},
 }

cmd/helmvm/join.go

@@ -13,49 +13,75 @@ import (
 
 	"github.com/replicatedhq/helmvm/pkg/defaults"
 	"github.com/replicatedhq/helmvm/pkg/goods"
+	"github.com/replicatedhq/helmvm/pkg/metrics"
 )
 
 var joinCommand = &cli.Command{
 	Name:  "join",
 	Usage: "Join the current node to an existing cluster",
-	Flags: []cli.Flag{
-		&cli.StringFlag{
-			Name:  "role",
-			Usage: "The role of the node (can be controller or worker)",
-			Value: "worker",
-		},
-	},
 	Action: func(c *cli.Context) error {
+		binname := defaults.BinaryName()
+		if c.Args().Len() != 1 {
+			return fmt.Errorf("usage: %s node join <token>", binname)
+		}
+		var hvmtoken JoinToken
+		if err := hvmtoken.Decode(c.Args().First()); err != nil {
+			return fmt.Errorf("unable to decode join token: %w", err)
+		}
+		metrics.ReportJoinStarted(c.Context, hvmtoken.ClusterID)
 		if err := canRunJoin(c); err != nil {
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
 			return err
 		}
 		logrus.Infof("Materializing binaries")
 		if err := goods.Materialize(); err != nil {
-			return fmt.Errorf("unable to materialize binaries: %w", err)
+			err := fmt.Errorf("unable to materialize binaries: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
 		}
 		if err := runHostPreflightsLocally(c); err != nil {
-			return fmt.Errorf("unable to run host preflights locally: %w", err)
+			err := fmt.Errorf("unable to run host preflights locally: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
 		}
 		logrus.Infof("Saving token to disk")
-		if err := saveTokenToDisk(c.Args().First()); err != nil {
-			return fmt.Errorf("unable to save token to disk: %w", err)
+		if err := saveTokenToDisk(hvmtoken.Token); err != nil {
+			err := fmt.Errorf("unable to save token to disk: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
 		}
 		logrus.Infof("Installing binary")
 		if err := installK0sBinary(); err != nil {
-			return fmt.Errorf("unable to install k0s binary: %w", err)
+			err := fmt.Errorf("unable to install k0s binary: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
 		}
 		logrus.Infof("Joining node to cluster")
-		if err := runK0sInstallCommand(c.String("role")); err != nil {
-			return fmt.Errorf("unable to join node to cluster: %w", err)
+		if err := runK0sInstallCommand(hvmtoken.Role); err != nil {
+			err := fmt.Errorf("unable to join node to cluster: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
 		}
 		logrus.Infof("Creating systemd unit file")
-		if err := createSystemdUnitFile(c.String("role")); err != nil {
-			return fmt.Errorf("unable to create systemd unit file: %w", err)
+		if err := createSystemdUnitFile(hvmtoken.Role); err != nil {
+			err := fmt.Errorf("unable to create systemd unit file: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
 		}
 		logrus.Infof("Starting service")
 		if err := startK0sService(); err != nil {
-			return fmt.Errorf("unable to start service: %w", err)
+			err := fmt.Errorf("unable to start service: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
 		}
+		fpath := defaults.PathToConfig(".cluster-id")
+		cid := hvmtoken.ClusterID.String()
+		if err := os.WriteFile(fpath, []byte(cid), 0644); err != nil {
+			err := fmt.Errorf("unable to write cluster id to disk: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
+		}
+		metrics.ReportJoinSucceeded(c.Context, hvmtoken.ClusterID)
 		return nil
 	},
 }
@@ -115,12 +141,6 @@ func canRunJoin(c *cli.Context) error {
 	if os.Getuid() != 0 {
 		return fmt.Errorf("join command must be run as root")
 	}
-	if c.Args().Len() != 1 {
-		return fmt.Errorf("usage: %s node join <token>", defaults.BinaryName())
-	}
-	if role := c.String("role"); role != "controller" && role != "worker" {
-		return fmt.Errorf("role must be either controller or worker")
-	}
 	return nil
 }
 

cmd/helmvm/token.go

@@ -2,17 +2,21 @@ package main
 
 import (
 	"bytes"
+	"encoding/base64"
+	"encoding/json"
 	"fmt"
 	"io"
 	"os"
 	"os/exec"
 	"runtime"
 	"time"
 
+	"github.com/google/uuid"
 	"github.com/sirupsen/logrus"
 	"github.com/urfave/cli/v2"
 
 	"github.com/replicatedhq/helmvm/pkg/defaults"
+	"github.com/replicatedhq/helmvm/pkg/metrics"
 	"github.com/replicatedhq/helmvm/pkg/prompts"
 )
 
@@ -22,6 +26,32 @@ var tokenCommands = &cli.Command{
 	Subcommands: []*cli.Command{tokenCreateCommand},
 }
 
+// JoinToken is a struct that holds both the actual token and the cluster id. This is marshaled
+// and base64 encoded and used as argument to the join command in the other nodes.
+type JoinToken struct {
+	ClusterID uuid.UUID `json:"clusterID"`
+	Token     string    `json:"token"`
+	Role      string    `json:"role"`
+}
+
+// Decode decodes a base64 encoded JoinToken.
+func (j *JoinToken) Decode(b64 string) error {
+	decoded, err := base64.StdEncoding.DecodeString(b64)
+	if err != nil {
+		return err
+	}
+	return json.Unmarshal(decoded, j)
+}
+
+// Encode encodes a JoinToken to base64.
+func (j *JoinToken) Encode() (string, error) {
+	b, err := json.Marshal(j)
+	if err != nil {
+		return "", err
+	}
+	return base64.StdEncoding.EncodeToString(b), nil
+}
+
 var tokenCreateCommand = &cli.Command{
 	Name:  "create",
 	Usage: "Creates a new node join token",
@@ -83,14 +113,19 @@ var tokenCreateCommand = &cli.Command{
 		}
 		if !defaults.DecentralizedInstall() {
 			if err := defaults.SetInstallAsDecentralized(); err != nil {
-				return fmt.Errorf("failed to set decentralized install: %w", err)
+				return fmt.Errorf("unable to set decentralized install: %w", err)
 			}
 		}
+		token := JoinToken{metrics.ClusterID(), buf.String(), role}
+		b64token, err := token.Encode()
+		if err != nil {
+			return fmt.Errorf("unable to encode token: %w", err)
+		}
 		fmt.Println("Token created successfully.")
 		fmt.Printf("This token is valid for %s hours.\n", dur)
 		fmt.Println("You can now run the following command in a remote node to add it")
 		fmt.Printf("to the cluster as a %q node:\n", role)
-		fmt.Printf("%s node join --role %s %s", defaults.BinaryName(), role, buf.String())
+		fmt.Printf("%s node join %s\n", defaults.BinaryName(), b64token)
 		return nil
 	},
 }

cmd/helmvm/upgrade.go

@@ -13,6 +13,7 @@ import (
 	"github.com/replicatedhq/helmvm/pkg/addons"
 	"github.com/replicatedhq/helmvm/pkg/defaults"
 	"github.com/replicatedhq/helmvm/pkg/goods"
+	"github.com/replicatedhq/helmvm/pkg/metrics"
 	"github.com/replicatedhq/helmvm/pkg/preflights"
 	"github.com/replicatedhq/helmvm/pkg/prompts"
 )
@@ -103,34 +104,49 @@ var upgradeCommand = &cli.Command{
 		},
 	},
 	Action: func(c *cli.Context) error {
+		metrics.ReportNodeUpgradeStarted(c.Context)
 		if err := canRunUpgrade(c); err != nil {
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
 			return err
 		}
 		logrus.Infof("Materializing binaries")
 		if err := goods.Materialize(); err != nil {
-			return fmt.Errorf("unable to materialize binaries: %w", err)
+			err := fmt.Errorf("unable to materialize binaries: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		if err := runHostPreflightsLocally(c); err != nil {
-			return fmt.Errorf("unable to run host preflights locally: %w", err)
+			err := fmt.Errorf("unable to run host preflights locally: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		logrus.Infof("Stopping %s", defaults.BinaryName())
 		if err := stopHelmVM(); err != nil {
-			return fmt.Errorf("unable to stop: %w", err)
+			err := fmt.Errorf("unable to stop: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		logrus.Infof("Installing binary")
 		if err := installK0sBinary(); err != nil {
-			return fmt.Errorf("unable to install k0s binary: %w", err)
+			err := fmt.Errorf("unable to install k0s binary: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		logrus.Infof("Starting service")
 		if err := startK0sService(); err != nil {
-			return fmt.Errorf("unable to start service: %w", err)
+			err := fmt.Errorf("unable to start service: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		kcfg := defaults.PathToConfig("kubeconfig")
 		if _, err := os.Stat(kcfg); err != nil {
 			if os.IsNotExist(err) {
+				metrics.ReportNodeUpgradeSucceeded(c.Context)
 				return nil
 			}
-			return fmt.Errorf("unable to stat kubeconfig: %w", err)
+			err := fmt.Errorf("unable to stat kubeconfig: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		os.Setenv("KUBECONFIG", kcfg)
 		logrus.Infof("Upgrading addons")
@@ -142,8 +158,11 @@ var upgradeCommand = &cli.Command{
 			opts = append(opts, addons.WithoutAddon(addon))
 		}
 		if err := addons.NewApplier(opts...).Apply(c.Context); err != nil {
-			return fmt.Errorf("unable to apply addons: %w", err)
+			err := fmt.Errorf("unable to apply addons: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
+		metrics.ReportNodeUpgradeSucceeded(c.Context)
 		logrus.Infof("Upgrade complete")
 		return nil
 	},

cmd/helmvm/version.go

@@ -11,8 +11,6 @@ import (
 	"github.com/replicatedhq/helmvm/pkg/defaults"
 )
 
-var Version = "v0.0.0"
-
 var versionCommand = &cli.Command{
 	Name:  "version",
 	Usage: fmt.Sprintf("Shows the %s installer version", defaults.BinaryName()),
@@ -24,7 +22,7 @@ var versionCommand = &cli.Command{
 		}
 		writer := table.NewWriter()
 		writer.AppendHeader(table.Row{"component", "version"})
-		writer.AppendRow(table.Row{"Installer", Version})
+		writer.AppendRow(table.Row{"Installer", defaults.Version})
 		writer.AppendRow(table.Row{"Kubernetes", defaults.K0sVersion})
 		for name, version := range versions {
 			if !strings.HasPrefix(version, "v") {

e2e/install_test.go

@@ -283,7 +283,7 @@ func TestHostPreflight(t *testing.T) {
 	defer tc.Destroy()
 	t.Log("installing ssh and binutils on node 0")
 	commands := [][]string{
-		{"dnf", "install", "-y", "openssh-server", "binutils", "tar"},
+		{"dnf", "--setopt=metadata_expire=120", "install", "-y", "openssh-server", "binutils", "tar"},
 		{"systemctl", "enable", "sshd"},
 		{"systemctl", "start", "sshd"},
 	}

e2e/utils.go

@@ -18,17 +18,22 @@ func (b *buffer) Close() error {
 	return nil
 }
 
+// RunCommandsOnNode runs a series of commands on a node.
 func RunCommandsOnNode(t *testing.T, cl *cluster.Output, node int, cmds [][]string) error {
 	for _, cmd := range cmds {
 		cmdstr := strings.Join(cmd, " ")
 		t.Logf("running `%s` node %d", cmdstr, node)
-		if _, _, err := RunCommandOnNode(t, cl, node, cmd); err != nil {
+		stdout, stderr, err := RunCommandOnNode(t, cl, node, cmd)
+		if err != nil {
+			t.Logf("stdout:\n%s", stdout)
+			t.Logf("stderr:\n%s", stderr)
 			return err
 		}
 	}
 	return nil
 }
 
+// RunCommandOnNode runs a command on a node with a timeout.
 func RunCommandOnNode(t *testing.T, cl *cluster.Output, node int, line []string) (string, string, error) {
 	stdout := &buffer{bytes.NewBuffer(nil)}
 	stderr := &buffer{bytes.NewBuffer(nil)}